Models & Labs
Kimi K3 vs GPT-5.6 Sol: Cost and Performance
Together AI Bloghigh confidence
Why it matters
- →Kimi K3 offers a cost-effective solution for high-volume tasks, outperforming Sol in multi-attempt scenarios.
- →GPT-5.6 Sol provides higher reliability in single-shot tasks, making it suitable for critical applications.
- →A combined routing strategy leverages both models' strengths, maximizing task coverage and efficiency.
Kimi K3 and GPT-5.6 Sol have been compared on the DeepSWE benchmark, revealing unique strengths for each model. GPT-5.6 Sol leads in single-attempt quality with a pass@1 score of 72.7%, while Kimi K3 excels in scenarios with multiple attempts, achieving a pass@4 score of 89.4%. Kimi K3 is also more cost-effective, providing 2.8 times more solved tasks per dollar than Sol. The models' differing strengths suggest that using a routing strategy, where tasks are first attempted by Kimi K3 and escalated to Sol if necessary, could optimize performance and cost-efficiency.
Read originalThe story around this
Earlier coverage that leads up to this article, and what followed. Lines connect each piece to the closest one after it, converging here.
How we got here
- 1OpenAI's GPT-5.6 Models Now in GitHub Copilot
GitHub Changelog · July 9, 2026 · Background
- 2OpenAI unveils GPT-5.6 model family
TechCrunch AI · July 9, 2026 · Background
- 3OpenAI Unveils GPT-5.6 with Sol Model
Fireship · July 10, 2026 · Background
- 4GPT-5.6 Sol Released as Practical Workhorse Model
The AI Daily Brief · July 10, 2026 · Background
- 5OpenAI Introduces Sol, Terra, and Luna Variants
Matt Wolfe · July 11, 2026 · Background
- 6Moonshot AI's Kimi K3 Challenges Frontier Models
The Rundown AI · July 17, 2026 · Related
- 7Kimi K3 Challenges Claude Fable 5 on DeepSWE
Together AI Blog · July 24, 2026 · Related
- 8Kimi K3 Open Weight Model Faces Scrutiny
Cole Medin · July 24, 2026 · Related
What happened next
- 9GPT-5.6 Triples Scores on ARC-AGI-3 Benchmark
OpenAI · July 29, 2026 · Background
- 10OpenAI Cuts Costs for GPT-5.6 Models
The Rundown AI · July 31, 2026 · Background
- 11DeepSeek V4 Pro 0813 vs GPT-5.6 Sol: Cost Efficiency
Together AI Blog · August 18, 2026 · Related
- 12GPT-6.1 Sol arrives in GitHub Copilot
GitHub Changelog · September 29, 2026 · Related
- 13OpenAI releases GPT-6.1 Sol, cuts costs while boosting safety
TechCrunch AI · September 29, 2026 · Related
More in Models & Labs
Models & Labsother
llama.cpp b12340 adds CUDA 13 and ROCm 10.0
This release quietly closes the hardware gap for local inference by adding default support for CUDA 13 and ROCm 10.0 across Linux and Windows. Users with newer NVIDIA GPUs or AMD accelerators no longer need to compile from source to get performance, while Apple Silicon builds now ship KleidiAI disabled by default, shifting the focus back to standard ARM optimizations. The inclusion of Snapdragon support for Linux and Android marks a significant step toward on-device AI beyond the x86/ARM desktop ecosystem. It is less about new features and more about ensuring llama.cpp remains the universal runtime for every available accelerator.
llama.cpp Releases
Models & Labscoding
llama.cpp b12342 adds few-row MMQ for speculative decoding
Speculative decoding just got a significant speed boost on NVIDIA GPUs. This release introduces a specialized kernel for 'few-row' matrix multiplications, targeting the small batch sizes typical of verification passes in speculative decoding workflows. By optimizing these specific operations for quantized formats like Q8_0 and IQ4 variants, llama.cpp reduces latency where it matters most during token generation. The change is CUDA-only for now but leverages hardware capabilities from Ampere onwards, making local inference faster without requiring new model architectures or complex configuration changes.
llama.cpp Releases
© TechCrunch AIModels & Labsmodels
Google deploys Gemini 4 Hydrogen for faster Search
Google is quietly shifting its search infrastructure by replacing older fast models with Gemini 4 Hydrogen, a tier designed specifically for speed and cost-efficiency rather than raw reasoning power. This model completes tasks 2.5 times faster than its predecessor while beating the more powerful Argon on specific intelligence benchmarks, proving that specialized efficiency can outperform brute force in high-volume workflows. By integrating this into Search's fan-out technique, Google now breaks down complex queries into subqueries with greater precision and speed, delivering richer visual responses without the latency of heavier models. This marks a strategic pivot toward agentic execution layers within consumer search, making AI Mode feel snappier for everyday multi-part questions.
TechCrunch AI